OpenClaw
A solo developer built an agent that actually does things, it went off like a firework, and six weeks later OpenAI hired the man and set the project free
Peter Steinberger built a thing called OpenClaw, and for a few weeks this winter it was the only thing anyone in my corner of the internet could talk about.
It's an agent that actually does things. Not a chatbot that advises you to book the flight — one that books it. Files the expense. Runs the errand. Wanders off into a little society of other agents and comes back having done the thing. "An agent even my mum can use," he called it, which is the highest bar in software and almost nobody clears it. It climbed GitHub faster than anything I can remember, renamed twice on the way up — once because Anthropic's lawyers felt the old name stood a little too close to Claude.
Then, maybe six weeks after it caught fire, OpenAI hired him. Not the company — there barely was one. The man. Altman said he'd "drive the next generation of personal agents." OpenClaw itself was set loose as an open foundation, free to keep living.
Sit with the shape of that. One builder, a pile of agents, six weeks from playground project to the most aggressive company in AI opening its chequebook — for a person, because a person who moves like that is the scarce thing now.
This is the small, senior team I've run my whole career, pushed to its limit: the team is one human and a fleet. And the prize everyone's chasing has quietly moved. Not the smartest model. The agent that acts — the one your mum can use.
"The claw is the law," he signed off. He's not entirely wrong.